| blob | 6650601c64f7644f612fa2488bdfd78a0a79720d |
1 /*
2 * Copyright (c) 2000-2006 Silicon Graphics, Inc.
3 * All Rights Reserved.
4 *
5 * This program is free software; you can redistribute it and/or
6 * modify it under the terms of the GNU General Public License as
7 * published by the Free Software Foundation.
8 *
9 * This program is distributed in the hope that it would be useful,
10 * but WITHOUT ANY WARRANTY; without even the implied warranty of
11 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
12 * GNU General Public License for more details.
13 *
14 * You should have received a copy of the GNU General Public License
15 * along with this program; if not, write the Free Software Foundation,
16 * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA
17 */
56 int
59 {
65 /*
66 * If it's a directory with any blocks, read-ahead block 0
67 * as we're almost certain to have the next operation be a read there.
68 */
74 }
76 }
78 /*
79 * xfs_getattr
80 */
81 int
86 {
105 #if XFS_BIG_INUMS
107 #endif
110 /*
111 * Quick exit for non-stat callers
112 */
118 /*
119 * Copy from in-core inode.
120 */
126 /*
127 * Check vnode type block/char vs. everything else.
128 */
142 /*
143 * If the file blocks are being allocated from a
144 * realtime partition, then return the inode's
145 * realtime extent size or the realtime volume's
146 * extent size.
147 */
150 }
152 }
160 /*
161 * Exit for stat callers. See if any of the rest of the fields
162 * to be filled in are needed.
163 */
169 /*
170 * Convert di_flags to xflags.
171 */
174 /*
175 * Exit for inode revalidate. See if any of the rest of
176 * the fields to be filled in are needed.
177 */
193 else
197 all_done:
201 }
204 /*
205 * xfs_setattr
206 */
207 int
213 {
219 uint lock_flags;
235 /*
236 * Cannot set certain attributes.
237 */
241 }
246 /*
247 * Timestamps do not need to be logged and hence do not
248 * need to be done within a transaction.
249 */
257 }
262 /*
263 * If disk quotas is on, we make sure that the dquots do exist on disk,
264 * before we start any other transactions. Trying to do this later
265 * is messy. We don't care to take a readlock to look at the ids
266 * in inode here, because we can't hold it across the trans_reserve.
267 * If the IDs do change before we take the ilock, we're covered
268 * because the i_*dquot fields will get updated anyway.
269 */
279 }
285 }
291 }
292 /*
293 * We take a reference when we initialize udqp and gdqp,
294 * so it is important that we never blindly double trip on
295 * the same variable. See xfs_create() for an example.
296 */
303 }
305 /*
306 * For the other attributes, we acquire the inode lock and
307 * first do an error checking pass.
308 */
323 }
324 }
334 }
335 }
338 }
342 /* boolean: are we the file owner? */
345 /*
346 * Change various properties of a file.
347 * Only the owner or users with CAP_FOWNER
348 * capability may do these things.
349 */
353 /*
354 * CAP_FOWNER overrides the following restrictions:
355 *
356 * The user ID of the calling process must be equal
357 * to the file owner ID, except in cases where the
358 * CAP_FSETID capability is applicable.
359 */
363 }
365 /*
366 * CAP_FSETID overrides the following restrictions:
367 *
368 * The effective user ID of the calling process shall match
369 * the file owner when setting the set-user-ID and
370 * set-group-ID bits on that file.
371 *
372 * The effective group ID or one of the supplementary group
373 * IDs of the calling process shall match the group owner of
374 * the file when setting the set-group-ID bit on that file
375 */
384 #if 0
385 /* Linux allows this, Irix doesn't. */
388 #endif
391 }
392 }
394 /*
395 * Change file ownership. Must be the owner or privileged.
396 * If the system was configured with the "restricted_chown"
397 * option, the owner is not permitted to give away the file,
398 * and can change the group id only to a group of which he
399 * or she is a member.
400 */
402 /*
403 * These IDs could have changed since we last looked at them.
404 * But, we're assured that if the ownership did change
405 * while we didn't have the inode locked, inode's dquot(s)
406 * would have changed also.
407 */
414 iprojid;
416 /*
417 * CAP_CHOWN overrides the following restrictions:
418 *
419 * If _POSIX_CHOWN_RESTRICTED is defined, this capability
420 * shall override the restriction that a process cannot
421 * change the user ID of a file it owns and the restriction
422 * that the group ID supplied to the chown() function
423 * shall be equal to either the group ID or one of the
424 * supplementary group IDs of the calling process.
425 */
432 }
433 /*
434 * Do a quota reservation only if uid/projid/gid is actually
435 * going to change.
436 */
446 }
447 }
449 /*
450 * Truncate file. Must have write permission and not be a directory.
451 */
453 /* Short circuit the truncate case for zero length files */
462 }
470 }
471 /*
472 * Make sure that the dquots are attached to the inode.
473 */
476 }
478 /*
479 * Change file access or modified times.
480 */
487 }
488 }
489 }
491 /*
492 * Change extent size or realtime flag.
493 */
495 /*
496 * Can't change extent size if any extents are allocated.
497 */
503 }
505 /*
506 * Can't change realtime flag if any extents are allocated.
507 */
514 }
515 /*
516 * Extent size must be a multiple of the appropriate block
517 * size, if set at all.
518 */
520 xfs_extlen_t size;
529 }
533 }
534 }
535 /*
536 * If realtime flag is set then must have realtime data.
537 */
545 }
546 }
548 /*
549 * Can't modify an immutable/append-only file unless
550 * we have appropriate permission.
551 */
560 }
561 }
563 /*
564 * Now we can make the changes. Before we join the inode
565 * to the transaction, if XFS_AT_SIZE is set then take care of
566 * the part of the truncation that must be done without the
567 * inode lock. This needs to be done before joining the inode
568 * to the transaction, because the inode cannot be unlocked
569 * once it is a part of the transaction.
570 */
576 }
579 /*
580 * We are going to log the inode size change in this
581 * transaction so any previous writes that are beyond the on
582 * disk EOF and the new EOF that have not been written out need
583 * to be written here. If we do not write the data out, we
584 * expose ourselves to the null files problem.
585 *
586 * Only flush from the on disk size to the smaller of the in
587 * memory file size or the new size as that's the range we
588 * really care about here and prevents waiting for other data
589 * not within the range we care about here.
590 */
597 }
599 /* wait for all I/O to complete */
609 }
613 XFS_TRANS_PERM_LOG_RES,
614 XFS_ITRUNCATE_LOG_COUNT))) {
619 }
622 }
627 }
629 /* determine whether mandatory locking mode changes */
632 /*
633 * Truncate file. Must have write permission and not be a directory.
634 */
636 /*
637 * Only change the c/mtime if we are changing the size
638 * or we are explicitly asked to change it. This handles
639 * the semantic difference between truncate() and ftruncate()
640 * as implemented in the VFS.
641 */
650 /*
651 * signal a sync transaction unless
652 * we're truncating an already unlinked
653 * file on a wsync filesystem
654 */
657 XFS_DATA_FORK,
663 /*
664 * Truncated "down", so we're removing references
665 * to old data here - if we now delay flushing for
666 * a long time, we expose ourselves unduly to the
667 * notorious NULL files problem. So, we mark this
668 * vnode and flush it when the file is closed, and
669 * do not wait the usual (long) time for writeout.
670 */
672 }
673 }
675 /*
676 * Change file access modes.
677 */
684 }
686 /*
687 * Change file ownership. Must be the owner or privileged.
688 * If the system was configured with the "restricted_chown"
689 * option, the owner is not permitted to give away the file,
690 * and can change the group id only to a group of which he
691 * or she is a member.
692 */
694 /*
695 * CAP_FSETID overrides the following restrictions:
696 *
697 * The set-user-ID and set-group-ID bits of a file will be
698 * cleared upon successful return from chown()
699 */
703 }
705 /*
706 * Change the ownerships and register quota modifications
707 * in the transaction.
708 */
715 }
717 }
725 }
727 }
735 }
737 /*
738 * We may have to rev the inode as well as
739 * the superblock version number since projids didn't
740 * exist before DINODE_VERSION_2 and SB_VERSION_NLINK.
741 */
744 }
748 }
751 /*
752 * Change file access or modified times.
753 */
760 }
766 }
769 }
771 /*
772 * Change XFS-added attributes.
773 */
776 /*
777 * Converting bytes to fs blocks.
778 */
781 }
783 uint di_flags;
785 /* can't set PREALLOC this way, just preserve it */
815 }
817 }
820 }
822 /*
823 * Change file inode change time only if XFS_AT_CTIME set
824 * AND we have been called by a DMI function.
825 */
832 }
834 /*
835 * Send out timestamp changes that need to be set to the
836 * current time. Not done when called by a DMI function.
837 */
843 /*
844 * If this is a synchronous mount, make sure that the
845 * transaction goes to disk before returning to the user.
846 * This is slightly sub-optimal in that truncates require
847 * two sync transactions instead of one for wsync filesystems.
848 * One for the truncate and one for the timestamps since we
849 * don't want to change the timestamps unless we're sure the
850 * truncate worked. Truncates are less than 1% of the laddis
851 * mix so this probably isn't worth the trouble to optimize.
852 */
859 }
861 /*
862 * If the (regular) file's mandatory locking mode changed, then
863 * notify the vnode. We do this under the inode lock to prevent
864 * racing calls to vop_vnode_change.
865 */
870 /*
871 * Release any dquot(s) the inode had kept before chown.
872 */
880 }
887 }
890 abort_return:
892 /* FALLTHROUGH */
893 error_return:
898 }
901 }
903 }
905 /*
906 * The maximum pathlen is 1024 bytes. Since the minimum file system
907 * blocksize is 512 bytes, we can get a max of 2 extents back from
908 * bmapi.
909 */
910 #define SYMLINK_MAPS 2
912 STATIC int
916 {
921 xfs_daddr_t d;
943 }
950 }
955 out:
957 }
959 int
963 {
987 }
989 out:
992 }
994 /*
995 * xfs_fsync
996 *
997 * This is called to sync the inode and its data out to disk.
998 * We need to hold the I/O lock while flushing the data, and
999 * the inode lock while flushing the inode. The inode lock CANNOT
1000 * be held while flushing the data, so acquire after we're done
1001 * with that.
1002 */
1003 int
1007 xfs_off_t start,
1008 xfs_off_t stop)
1009 {
1024 /*
1025 * We always need to make sure that the required inode state
1026 * is safe on disk. The vnode might be clean but because
1027 * of committed transactions that haven't hit the disk yet.
1028 * Likewise, there could be unflushed non-transactional
1029 * changes to the inode core that have to go to disk.
1030 *
1031 * The following code depends on one assumption: that
1032 * any transaction that changes an inode logs the core
1033 * because it has to change some field in the inode core
1034 * (typically nextents or nblocks). That assumption
1035 * implies that any transactions against an inode will
1036 * catch any non-transactional updates. If inode-altering
1037 * transactions exist that violate this assumption, the
1038 * code breaks. Right now, it figures that if the involved
1039 * update_* field is clear and the inode is unpinned, the
1040 * inode is clean. Either it's been flushed or it's been
1041 * committed and the commit has hit the disk unpinning the inode.
1042 * (Note that xfs_inode_item_format() called at commit clears
1043 * the update_* fields.)
1044 */
1047 /* If we are flushing data then we care about update_size
1048 * being set, otherwise we care about update_core
1049 */
1053 /*
1054 * Timestamps/size haven't changed since last inode
1055 * flush or inode transaction commit. That means
1056 * either nothing got written or a transaction
1057 * committed which caught the updates. If the
1058 * latter happened and the transaction hasn't
1059 * hit the disk yet, the inode will be still
1060 * be pinned. If it is, force the log.
1061 */
1067 XFS_LOG_FORCE |
1072 /*
1073 * If the inode is not pinned and nothing
1074 * has changed we don't need to flush the
1075 * cache.
1076 */
1078 }
1081 /*
1082 * Kick off a transaction to log the inode
1083 * core to get the updates. Make it
1084 * sync if FSYNC_WAIT is passed in (which
1085 * is done by everybody but specfs). The
1086 * sync transaction will also force the log.
1087 */
1095 }
1098 /*
1099 * Note - it's possible that we might have pushed
1100 * ourselves out of the way during trans_reserve
1101 * which would flush the inode. But there's no
1102 * guarantee that the inode buffer has actually
1103 * gone out yet (it's delwri). Plus the buffer
1104 * could be pinned anyway if it's part of an
1105 * inode in another recent transaction. So we
1106 * play it safe and fire off the transaction anyway.
1107 */
1116 }
1119 /*
1120 * If the log write didn't issue an ordered tag we need
1121 * to flush the disk cache for the data device now.
1122 */
1126 /*
1127 * If this inode is on the RT dev we need to flush that
1128 * cache as well.
1129 */
1132 }
1135 }
1137 /*
1138 * This is called by xfs_inactive to free any blocks beyond eof
1139 * when the link count isn't zero and by xfs_dm_punch_hole() when
1140 * punching a hole to EOF.
1141 */
1142 int
1147 {
1150 xfs_fileoff_t end_fsb;
1151 xfs_fileoff_t last_fsb;
1152 xfs_filblks_t map_len;
1154 xfs_bmbt_irec_t imap;
1157 /*
1158 * Figure out if there are any blocks beyond the end
1159 * of the file. If not, then there is nothing to do.
1160 */
1176 /*
1177 * Attach the dquots to the inode up front.
1178 */
1182 /*
1183 * There are blocks after the end of file.
1184 * Free them up now by truncating the file to
1185 * its current size.
1186 */
1189 /*
1190 * Do the xfs_itruncate_start() call before
1191 * reserving any log space because
1192 * itruncate_start will call into the buffer
1193 * cache and we can't
1194 * do that within a transaction.
1195 */
1205 }
1210 XFS_ITRUNCATE_LOG_COUNT);
1216 }
1220 XFS_IOLOCK_EXCL |
1221 XFS_ILOCK_EXCL);
1226 XFS_DATA_FORK,
1228 /*
1229 * If we get an error at this point we
1230 * simply don't bother truncating the file.
1231 */
1235 XFS_TRANS_ABORT));
1238 XFS_TRANS_RELEASE_LOG_RES);
1239 }
1242 }
1244 }
1246 /*
1247 * Free a symlink that has blocks associated with it.
1248 */
1249 STATIC int
1253 {
1258 xfs_fsblock_t first_block;
1259 xfs_bmap_free_t free_list;
1271 /*
1272 * We're freeing a symlink that has some
1273 * blocks allocated to it. Free the
1274 * blocks here. We know that we've got
1275 * either 1 or 2 extents and that we can
1276 * free them all in one bunmapi call.
1277 */
1285 }
1286 /*
1287 * Lock the inode, fix the size, and join it to the transaction.
1288 * Hold it so in the normal path, we still have it locked for
1289 * the second transaction. In the error paths we need it
1290 * held so the cancel won't rele it, see below.
1291 */
1298 /*
1299 * Find the block(s) so we can inval and unmap them.
1300 */
1308 /*
1309 * Invalidate the block(s).
1310 */
1316 }
1317 /*
1318 * Unmap the dead block(s) to the free_list.
1319 */
1324 /*
1325 * Commit the first transaction. This logs the EFI and the inode.
1326 */
1329 /*
1330 * The transaction must have been committed, since there were
1331 * actually extents freed by xfs_bunmapi. See xfs_bmap_finish.
1332 * The new tp has the extent freeing and EFDs.
1333 */
1335 /*
1336 * The first xact was committed, so add the inode to the new one.
1337 * Mark it dirty so it will be logged and moved forward in the log as
1338 * part of every commit.
1339 */
1343 /*
1344 * Get a new, empty transaction to return to our caller.
1345 */
1347 /*
1348 * Commit the transaction containing extent freeing and EFDs.
1349 * If we get an error on the commit here or on the reserve below,
1350 * we need to unlock the inode since the new transaction doesn't
1351 * have the inode attached.
1352 */
1358 }
1359 /*
1360 * Remove the memory for extent descriptions (just bookkeeping).
1361 */
1365 /*
1366 * Put an itruncate log reservation in the new transaction
1367 * for our caller.
1368 */
1373 }
1374 /*
1375 * Return with the inode locked but not joined to the transaction.
1376 */
1380 error1:
1382 error0:
1383 /*
1384 * Have to come here with the inode locked and either
1385 * (held and in the transaction) or (not in the transaction).
1386 * If the inode isn't held then cancel would iput it, but
1387 * that's wrong since this is inactive and the vnode ref
1388 * count is 0 already.
1389 * Cancel won't do anything to the inode if held, but it still
1390 * needs to be locked until the cancel is done, if it was
1391 * joined to the transaction.
1392 */
1398 }
1400 STATIC int
1404 {
1408 /*
1409 * We're freeing a symlink which fit into
1410 * the inode. Just free the memory used
1411 * to hold the old symlink.
1412 */
1416 XFS_ITRUNCATE_LOG_COUNT);
1422 }
1425 /*
1426 * Zero length symlinks _can_ exist.
1427 */
1431 XFS_DATA_FORK);
1433 }
1435 }
1437 STATIC int
1441 {
1463 XFS_INACTIVE_LOG_COUNT);
1477 error_cancel:
1480 error_unlock:
1484 }
1486 int
1489 {
1497 /* If this is a read-only mount, don't do this (would generate I/O) */
1504 /*
1505 * If we are using filestreams, and we have an unlinked
1506 * file that we are processing the last close on, then nothing
1507 * will be able to reopen and write to this file. Purge this
1508 * inode from the filestreams cache so that it doesn't delay
1509 * teardown of the inode.
1510 */
1514 /*
1515 * If we previously truncated this file and removed old data
1516 * in the process, we want to initiate "early" writeout on
1517 * the last close. This is an attempt to combat the notorious
1518 * NULL files problem which is particularly noticable from a
1519 * truncate down, buffered (re-)write (delalloc), followed by
1520 * a crash. What we are effectively doing here is
1521 * significantly reducing the time window where we'd otherwise
1522 * be exposed to that problem.
1523 */
1527 }
1539 }
1540 }
1543 }
1545 /*
1546 * xfs_inactive
1547 *
1548 * This is called when the vnode reference count for the vnode
1549 * goes to zero. If the file has been unlinked, then it must
1550 * now be truncated. Also, we clear all of the read-ahead state
1551 * kept for the inode here since the file is now closed.
1552 */
1553 int
1556 {
1558 xfs_bmap_free_t free_list;
1559 xfs_fsblock_t first_block;
1568 /*
1569 * If the inode is already free, then there can be nothing
1570 * to clean up here.
1571 */
1576 }
1578 /*
1579 * Only do a truncate if it's a regular file with
1580 * some actual space in it. It's OK to look at the
1581 * inode's fields without the lock because we're the
1582 * only one with a reference to the inode.
1583 */
1596 /* If this is a read-only mount, don't do this (would generate I/O) */
1611 }
1613 }
1622 /*
1623 * Do the xfs_itruncate_start() call before
1624 * reserving any log space because itruncate_start
1625 * will call into the buffer cache and we can't
1626 * do that within a transaction.
1627 */
1635 }
1640 XFS_ITRUNCATE_LOG_COUNT);
1642 /* Don't call itruncate_cleanup */
1647 }
1653 /*
1654 * normally, we have to run xfs_itruncate_finish sync.
1655 * But if filesystem is wsync and we're in the inactive
1656 * path, then we know that nlink == 0, and that the
1657 * xaction that made nlink == 0 is permanently committed
1658 * since xfs_remove runs as a synchronous transaction.
1659 */
1668 }
1671 /*
1672 * If we get an error while cleaning up a
1673 * symlink we bail out.
1674 */
1682 }
1690 XFS_INACTIVE_LOG_COUNT);
1695 }
1700 }
1702 /*
1703 * If there are attributes associated with the file
1704 * then blow them away now. The code calls a routine
1705 * that recursively deconstructs the attribute fork.
1706 * We need to just commit the current transaction
1707 * because we can't use it for xfs_attr_inactive().
1708 */
1711 /*
1712 * If we got an error, the transaction is already
1713 * cancelled, and the inode is unlocked. Just get out.
1714 */
1719 }
1721 /*
1722 * Free the inode.
1723 */
1727 /*
1728 * If we fail to free the inode, shut down. The cancel
1729 * might do that, we need to make sure. Otherwise the
1730 * inode might be lost for a long time or forever.
1731 */
1737 }
1740 /*
1741 * Credit the quota account(s). The inode is gone.
1742 */
1745 /*
1746 * Just ignore errors at this point. There is nothing we can
1747 * do except to try to keep going. Make sure it's not a silent
1748 * error.
1749 */
1758 }
1759 /*
1760 * Release the dquots held by inode, if any.
1761 */
1766 out:
1768 }
1771 int
1776 {
1778 xfs_ino_t e_inum;
1780 uint lock_mode;
1792 }
1795 }
1797 int
1801 mode_t mode,
1802 xfs_dev_t rdev,
1805 {
1810 xfs_bmap_free_t free_list;
1811 xfs_fsblock_t first_block;
1814 uint cancel_flags;
1816 xfs_prid_t prid;
1818 uint resblks;
1832 }
1837 /* Return through std_return after this point. */
1842 else
1845 /*
1846 * Make sure that we have allocated dquot(s) on disk.
1847 */
1859 /*
1860 * Initially assume that the file does not exist and
1861 * reserve the resources for that case. If that is not
1862 * the case we'll drop the one we have and get a more
1863 * appropriate transaction later.
1864 */
1871 }
1875 }
1884 /*
1885 * Reserve disk quota and the inode.
1886 */
1901 }
1904 /*
1905 * At this point, we've gotten a newly allocated inode.
1906 * It is locked (and joined to the transaction).
1907 */
1911 /*
1912 * Now we join the directory inode to the transaction. We do not do it
1913 * earlier because xfs_dir_ialloc might commit the previous transaction
1914 * (and release all the locks). An error from here on will result in
1915 * the transaction cancel unlocking dp so don't do it explicitly in the
1916 * error path.
1917 */
1928 }
1932 /*
1933 * If this is a synchronous mount, make sure that the
1934 * create transaction goes to disk before returning to
1935 * the user.
1936 */
1939 }
1943 /*
1944 * Attach the dquot(s) to the inodes and modify them incore.
1945 * These ids of the inode couldn't have changed since the new
1946 * inode has been locked ever since it was created.
1947 */
1950 /*
1951 * xfs_trans_commit normally decrements the vnode ref count
1952 * when it unlocks the inode. Since we want to return the
1953 * vnode to the caller, we bump the vnode ref count now.
1954 */
1961 }
1968 }
1975 /* Fallthrough to std_return with error = 0 */
1977 std_return:
1985 }
1988 abort_return:
1990 /* FALLTHROUGH */
1992 error_return:
2004 abort_rele:
2005 /*
2006 * Wait until after the current transaction is aborted to
2007 * release the inode. This prevents recursive transactions
2008 * and deadlocks from xfs_inactive.
2009 */
2018 }
2020 #ifdef DEBUG
2021 /*
2022 * Some counters to see if (and how often) we are hitting some deadlock
2023 * prevention code paths.
2024 */
2029 #endif
2031 /*
2032 * The following routine will lock the inodes associated with the
2033 * directory and the named entry in the directory. The locks are
2034 * acquired in increasing inode number.
2035 *
2036 * If the entry is "..", then only the directory is locked. The
2037 * vnode ref count will still include that from the .. entry in
2038 * this case.
2039 *
2040 * There is a deadlock we need to worry about. If the locked directory is
2041 * in the AIL, it might be blocking up the log. The next inode we lock
2042 * could be already locked by another thread waiting for log space (e.g
2043 * a permanent log reservation with a long running transaction (see
2044 * xfs_itruncate_finish)). To solve this, we must check if the directory
2045 * is in the ail and use lock_nowait. If we can't lock, we need to
2046 * drop the inode lock on the directory and try again. xfs_iunlock will
2047 * potentially push the tail if we were holding up the log.
2048 */
2049 STATIC int
2053 {
2055 xfs_ino_t e_inum;
2059 #ifdef DEBUG
2060 xfs_rm_locks++;
2061 #endif
2064 again:
2071 /*
2072 * We want to lock in increasing inum. Since we've already
2073 * acquired the lock on the directory, we may need to release
2074 * if if the inum of the entry turns out to be less.
2075 */
2077 /*
2078 * We are already in the right order, so just
2079 * lock on the inode of the entry.
2080 * We need to use nowait if dp is in the AIL.
2081 */
2086 attempts++;
2087 #ifdef DEBUG
2088 xfs_rm_attempts++;
2089 #endif
2091 /*
2092 * Unlock dp and try again.
2093 * xfs_iunlock will try to push the tail
2094 * if the inode is in the AIL.
2095 */
2101 #ifdef DEBUG
2102 xfs_rm_lock_delays++;
2103 #endif
2104 }
2106 }
2109 }
2116 }
2117 /* else e_inum == dp->i_ino */
2118 /* This can happen if we're asked to lock /x/..
2119 * the entry is "..", which is also the parent directory.
2120 */
2123 }
2125 #ifdef DEBUG
2131 #endif
2133 /*
2134 * Bump the subclass so xfs_lock_inodes() acquires each lock with
2135 * a different value
2136 */
2139 {
2146 }
2148 /*
2149 * The following routine will lock n inodes in exclusive mode.
2150 * We assume the caller calls us with the inodes in i_ino order.
2151 *
2152 * We need to detect deadlock where an inode that we lock
2153 * is in the AIL and we start waiting for another inode that is locked
2154 * by a thread in a long running transaction (such as truncate). This can
2155 * result in deadlock since the long running trans might need to wait
2156 * for the inode we just locked in order to push the tail and free space
2157 * in the log.
2158 */
2159 void
2164 uint lock_mode)
2165 {
2177 }
2179 again:
2186 /*
2187 * If try_lock is not set yet, make sure all locked inodes
2188 * are not in the AIL.
2189 * If any are, set try_lock to be used later.
2190 */
2196 try_lock++;
2197 }
2198 }
2199 }
2201 /*
2202 * If any of the previous locks we have locked is in the AIL,
2203 * we must TRY to get the second and subsequent locks. If
2204 * we can't get any, we must release all we have
2205 * and try again.
2206 */
2209 /* try_lock must be 0 if i is 0. */
2210 /*
2211 * try_lock means we have an inode locked
2212 * that is in the AIL.
2213 */
2216 attempts++;
2218 /*
2219 * Unlock all previous guys and try again.
2220 * xfs_iunlock will try to push the tail
2221 * if the inode is in the AIL.
2222 */
2226 /*
2227 * Check to see if we've already
2228 * unlocked this one.
2229 * Not the first one going back,
2230 * and the inode ptr is the same.
2231 */
2237 }
2241 #ifdef DEBUG
2242 xfs_lock_delays++;
2243 #endif
2244 }
2248 }
2251 }
2252 }
2254 #ifdef DEBUG
2260 xfs_locked_n++;
2261 }
2262 #endif
2263 }
2265 #ifdef DEBUG
2266 #define REMOVE_DEBUG_TRACE(x) {remove_which_error_return = (x);}
2269 #define REMOVE_DEBUG_TRACE(x)
2272 int
2277 {
2281 xfs_bmap_free_t free_list;
2282 xfs_fsblock_t first_block;
2286 uint resblks;
2299 }
2301 /*
2302 * We need to get a reference to ip before we get our log
2303 * reservation. The reason for this is that we cannot call
2304 * xfs_iget for an inode for which we do not have a reference
2305 * once we've acquired a log reservation. This is because the
2306 * inode we are trying to get might be in xfs_inactive going
2307 * for a log reservation. Since we'll have to wait for the
2308 * inactive code to complete before returning from xfs_iget,
2309 * we need to make sure that we don't have log space reserved
2310 * when we call xfs_iget. Instead we get an unlocked reference
2311 * to the inode before getting our log reservation.
2312 */
2325 }
2329 /*
2330 * We try to get the real space reservation first,
2331 * allowing for directory btree deletion(s) implying
2332 * possible bmap insert(s). If we can't get the space
2333 * reservation then we use 0 instead, and avoid the bmap
2334 * btree insert(s) in the directory code by, if the bmap
2335 * insert tries to happen, instead trimming the LAST
2336 * block from the directory.
2337 */
2345 }
2352 }
2360 }
2362 /*
2363 * At this point, we've gotten both the directory and the entry
2364 * inodes locked.
2365 */
2368 /*
2369 * Increment vnode ref count only in this case since
2370 * there's an extra vnode reference in the case where
2371 * dp == ip.
2372 */
2375 }
2377 /*
2378 * Entry must exist since we did a lookup in xfs_lock_dir_and_entry.
2379 */
2387 }
2397 }
2399 /* Determine if this is the last link while
2400 * we are in the transaction.
2401 */
2404 /*
2405 * Take an extra ref on the inode so that it doesn't
2406 * go to xfs_inactive() from within the commit.
2407 */
2410 /*
2411 * If this is a synchronous mount, make sure that the
2412 * remove transaction goes to disk before returning to
2413 * the user.
2414 */
2417 }
2423 }
2429 }
2431 /*
2432 * If we are using filestreams, kill the stream association.
2433 * If the file is still open it may get a new one but that
2434 * will get killed on last close in xfs_close() so we don't
2435 * have to worry about that.
2436 */
2443 /* Fall through to std_return with error = 0 */
2444 std_return:
2450 }
2453 error1:
2459 error_rele:
2460 /*
2461 * In this case make sure to not release the inode until after
2462 * the current transaction is aborted. Releasing it beforehand
2463 * can cause us to go to xfs_inactive and start a recursive
2464 * transaction which can easily deadlock with the current one.
2465 */
2473 }
2475 int
2480 {
2485 xfs_bmap_free_t free_list;
2486 xfs_fsblock_t first_block;
2506 }
2508 /* Return through std_return after this point. */
2525 }
2529 }
2537 }
2541 /*
2542 * Increment vnode ref counts since xfs_trans_commit &
2543 * xfs_trans_cancel will both unlock the inodes and
2544 * decrement the associated ref counts.
2545 */
2551 /*
2552 * If the source has too many links, we can't make any more to it.
2553 */
2557 }
2559 /*
2560 * If we are using project inheritance, we only allow hard link
2561 * creation in our tree when the project IDs are the same; else
2562 * the tree quota mechanism could be circumvented.
2563 */
2568 }
2588 /*
2589 * If this is a synchronous mount, make sure that the
2590 * link transaction goes to disk before returning to
2591 * the user.
2592 */
2595 }
2601 }
2607 /* Fall through to std_return with error = 0. */
2608 std_return:
2614 }
2617 abort_return:
2619 /* FALLTHROUGH */
2621 error_return:
2624 }
2627 int
2631 mode_t mode,
2634 {
2641 xfs_bmap_free_t free_list;
2642 xfs_fsblock_t first_block;
2646 xfs_prid_t prid;
2648 uint resblks;
2663 }
2665 /* Return through std_return after this point. */
2673 else
2676 /*
2677 * Make sure that we have allocated dquot(s) on disk.
2678 */
2693 XFS_TRANS_PERM_LOG_RES,
2694 XFS_MKDIR_LOG_COUNT);
2695 }
2699 }
2704 /*
2705 * Check for directory link count overflow.
2706 */
2710 }
2712 /*
2713 * Reserve disk quota and the inode.
2714 */
2722 /*
2723 * create the directory inode.
2724 */
2732 }
2735 /*
2736 * Now we add the directory inode to the transaction.
2737 * We waited until now since xfs_dir_ialloc might start
2738 * a new transaction. Had we joined the transaction
2739 * earlier, the locks might have gotten released. An error
2740 * from here on will result in the transaction cancel
2741 * unlocking dp so don't do it explicitly in the error path.
2742 */
2755 }
2758 /*
2759 * Bump the in memory version number of the parent directory
2760 * so that other processes accessing it will recognize that
2761 * the directory has changed.
2762 */
2779 /*
2780 * Attach the dquots to the new inode and modify the icount incore.
2781 */
2784 /*
2785 * If this is a synchronous mount, make sure that the
2786 * mkdir transaction goes to disk before returning to
2787 * the user.
2788 */
2791 }
2797 }
2804 }
2806 /* Fall through to std_return with error = 0 or errno from
2807 * xfs_trans_commit. */
2809 std_return:
2815 DM_RIGHT_NULL,
2818 }
2821 error2:
2822 error1:
2824 abort_return:
2826 error_return:
2835 }
2837 int
2842 {
2847 xfs_bmap_free_t free_list;
2848 xfs_fsblock_t first_block;
2852 uint resblks;
2866 }
2868 /*
2869 * We need to get a reference to cdp before we get our log
2870 * reservation. The reason for this is that we cannot call
2871 * xfs_iget for an inode for which we do not have a reference
2872 * once we've acquired a log reservation. This is because the
2873 * inode we are trying to get might be in xfs_inactive going
2874 * for a log reservation. Since we'll have to wait for the
2875 * inactive code to complete before returning from xfs_iget,
2876 * we need to make sure that we don't have log space reserved
2877 * when we call xfs_iget. Instead we get an unlocked reference
2878 * to the inode before getting our log reservation.
2879 */
2882 /*
2883 * Get the dquots for the inodes.
2884 */
2892 }
2896 /*
2897 * We try to get the real space reservation first,
2898 * allowing for directory btree deletion(s) implying
2899 * possible bmap insert(s). If we can't get the space
2900 * reservation then we use 0 instead, and avoid the bmap
2901 * btree insert(s) in the directory code by, if the bmap
2902 * insert tries to happen, instead trimming the LAST
2903 * block from the directory.
2904 */
2912 }
2918 }
2921 /*
2922 * Now lock the child directory inode and the parent directory
2923 * inode in the proper order. This will take care of validating
2924 * that the directory entry for the child directory inode has
2925 * not changed while we were obtaining a log reservation.
2926 */
2932 }
2936 /*
2937 * Only increment the parent directory vnode count if
2938 * we didn't bump it in looking up cdp. The only time
2939 * we don't bump it is when we're looking up ".".
2940 */
2942 }
2951 }
2955 }
2964 /*
2965 * Bump the in memory generation count on the parent
2966 * directory so that other can know that it has changed.
2967 */
2970 /*
2971 * Drop the link from cdp's "..".
2972 */
2976 }
2978 /*
2979 * Drop the link from dp to cdp.
2980 */
2984 }
2986 /*
2987 * Drop the "." link from cdp to self.
2988 */
2992 }
2994 /* Determine these before committing transaction */
2997 /*
2998 * Take an extra ref on the child vnode so that it
2999 * does not go to xfs_inactive() from within the commit.
3000 */
3003 /*
3004 * If this is a synchronous mount, make sure that the
3005 * rmdir transaction goes to disk before returning to
3006 * the user.
3007 */
3010 }
3016 XFS_TRANS_ABORT));
3019 }
3025 }
3030 /* Fall through to std_return with error = 0 or the errno
3031 * from xfs_trans_commit. */
3032 std_return:
3039 }
3042 error1:
3045 /* FALLTHROUGH */
3047 error_return:
3050 }
3052 int
3057 mode_t mode,
3060 {
3066 xfs_bmap_free_t free_list;
3067 xfs_fsblock_t first_block;
3069 uint cancel_flags;
3071 xfs_fileoff_t first_fsb;
3072 xfs_filblks_t fs_blocks;
3075 xfs_daddr_t d;
3080 xfs_prid_t prid;
3082 uint resblks;
3094 /*
3095 * Check component lengths of the target path name.
3096 */
3107 }
3109 /* Return through std_return after this point. */
3114 else
3117 /*
3118 * Make sure that we have allocated dquot(s) on disk.
3119 */
3128 /*
3129 * The symlink will fit into the inode data fork?
3130 * There can't be any attributes so we get the whole variable part.
3131 */
3134 else
3143 }
3147 }
3152 /*
3153 * Check whether the directory allows new symlinks or not.
3154 */
3158 }
3160 /*
3161 * Reserve disk quota : blocks and inode.
3162 */
3167 /*
3168 * Check for ability to enter directory entry, if no space reserved.
3169 */
3173 /*
3174 * Initialize the bmap freelist prior to calling either
3175 * bmapi or the directory create code.
3176 */
3179 /*
3180 * Allocate an inode for the symlink.
3181 */
3188 }
3191 /*
3192 * An error after we've joined dp to the transaction will result in the
3193 * transaction cancel unlocking dp so don't do it explicitly in the
3194 * error path.
3195 */
3200 /*
3201 * Also attach the dquot(s) to it, if applicable.
3202 */
3207 /*
3208 * If the symlink will fit into the inode, write it inline.
3209 */
3215 /*
3216 * The inode was initially created in extent format.
3217 */
3234 }
3250 }
3257 }
3258 }
3260 /*
3261 * Create the directory entry for the symlink.
3262 */
3270 /*
3271 * Bump the in memory version number of the parent directory
3272 * so that other processes accessing it will recognize that
3273 * the directory has changed.
3274 */
3277 /*
3278 * If this is a synchronous mount, make sure that the
3279 * symlink transaction goes to disk before returning to
3280 * the user.
3281 */
3284 }
3286 /*
3287 * xfs_trans_commit normally decrements the vnode ref count
3288 * when it unlocks the inode. Since we want to return the
3289 * vnode to the caller, we bump the vnode ref count now.
3290 */
3296 }
3301 /* Fall through to std_return with error = 0 or errno from
3302 * xfs_trans_commit */
3303 std_return:
3310 }
3316 error2:
3318 error1:
3321 error_return:
3330 }
3332 int
3336 {
3343 /*
3344 * Bypass inodes which have already been cleaned by
3345 * the inode flush clustering code inside xfs_iflush
3346 */
3350 /*
3351 * We make this non-blocking if the inode is contended,
3352 * return EAGAIN to indicate to the caller that they
3353 * did not succeed. This prevents the flush path from
3354 * blocking on inodes inside another operation right
3355 * now, they get caught later by xfs_sync.
3356 */
3364 }
3367 }
3374 }
3377 int
3380 u_int evmask,
3381 u_int16_t state)
3382 {
3398 }
3410 }
3412 int
3415 {
3422 /* bad inode, get out here ASAP */
3426 }
3432 /*
3433 * Make sure the atime in the XFS inode is correct before freeing the
3434 * Linux inode.
3435 */
3438 /*
3439 * If we have nothing to flush with this inode then complete the
3440 * teardown now, otherwise break the link between the xfs inode and the
3441 * linux inode and clean up the xfs inode later. This avoids flushing
3442 * the inode to disk during the delete operation itself.
3443 *
3444 * When breaking the link, we need to set the XFS_IRECLAIMABLE flag
3445 * first to ensure that xfs_iunpin() will never see an xfs inode
3446 * that has a linux inode being reclaimed. Synchronisation is provided
3447 * by the i_flags_lock.
3448 */
3456 /* Protect sync and unpin from us */
3465 }
3467 }
3469 int
3474 {
3482 /* The hash lock here protects a thread in xfs_iget_core from
3483 * racing with us on linking the inode back with a vnode.
3484 * Once we have the XFS_IRECLAIM flag set it will not touch
3485 * us.
3486 */
3496 }
3498 }
3504 /*
3505 * If the inode is still dirty, then flush it out. If the inode
3506 * is not in the AIL, then it will be OK to flush it delwri as
3507 * long as xfs_iflush() does not keep any references to the inode.
3508 * We leave that decision up to xfs_iflush() since it has the
3509 * knowledge of whether it's OK to simply do a delwri flush of
3510 * the inode or whether we need to wait until the inode is
3511 * pulled from the AIL.
3512 * We get the flush lock regardless, though, just to make sure
3513 * we don't free it while it is being flushed.
3514 */
3518 }
3525 /*
3526 * If we hit an error, typically because of filesystem
3527 * shutdown, we don't need to let vn_reclaim to know
3528 * because we're gonna reclaim the inode anyway.
3529 */
3533 }
3535 }
3540 }
3545 reclaim:
3548 }
3550 int
3552 {
3568 }
3569 }
3572 XFS_IFLUSH_DELWRI_ELSE_ASYNC))
3576 }
3579 }
3583 }
3585 /*
3586 * xfs_alloc_file_space()
3587 * This routine allocates disk space for the given file.
3588 *
3589 * If alloc_type == 0, this request is for an ALLOCSP type
3590 * request which will change the file size. In this case, no
3591 * DMAPI event will be generated by the call. A TRUNCATE event
3592 * will be generated later by xfs_setattr.
3593 *
3594 * If alloc_type != 0, this request is for a RESVSP type
3595 * request, and a DMAPI DM_EVENT_WRITE will be generated if the
3596 * lower block boundary byte address is less than the file's
3597 * length.
3598 *
3599 * RETURNS:
3600 * 0 on success
3601 * errno on error
3602 *
3603 */
3604 STATIC int
3607 xfs_off_t offset,
3608 xfs_off_t len,
3611 {
3613 xfs_off_t count;
3614 xfs_filblks_t allocated_fsb;
3615 xfs_filblks_t allocatesize_fsb;
3617 xfs_fileoff_t startoffset_fsb;
3618 xfs_fsblock_t firstfsb;
3625 xfs_bmap_free_t free_list;
3651 /* Generate a DMAPI event if needed. */
3655 xfs_off_t end_dmi_offset;
3664 }
3666 /*
3667 * Allocate file space until done or until there is an error
3668 */
3669 retry:
3673 /*
3674 * Determine space reservations for data/realtime.
3675 */
3688 }
3700 }
3702 /*
3703 * Allocate and setup the transaction.
3704 */
3708 XFS_TRANS_PERM_LOG_RES,
3709 XFS_WRITE_LOG_COUNT);
3710 /*
3711 * Check for running out of space
3712 */
3714 /*
3715 * Free the transaction structure.
3716 */
3720 }
3730 /*
3731 * Issue the xfs_bmapi() call to allocate the blocks
3732 */
3740 }
3742 /*
3743 * Complete the transaction
3744 */
3748 }
3754 }
3761 }
3765 }
3766 dmapi_enospc_check:
3775 /* else fall through with error from XFS_SEND_DATA */
3776 }
3788 }
3790 /*
3791 * Zero file bytes between startoff and endoff inclusive.
3792 * The iolock is held exclusive and no blocks are buffered.
3793 */
3794 STATIC int
3797 xfs_off_t startoff,
3798 xfs_off_t endoff)
3799 {
3800 xfs_bmbt_irec_t imap;
3801 xfs_fileoff_t offset_fsb;
3802 xfs_off_t lastoffset;
3803 xfs_off_t offset;
3840 }
3853 }
3854 }
3857 }
3859 /*
3860 * xfs_free_file_space()
3861 * This routine frees disk space for the given file.
3862 *
3863 * This routine is only called by xfs_change_file_space
3864 * for an UNRESVSP type call.
3865 *
3866 * RETURNS:
3867 * 0 on success
3868 * errno on error
3869 *
3870 */
3871 STATIC int
3874 xfs_off_t offset,
3875 xfs_off_t len,
3877 {
3881 xfs_off_t end_dmi_offset;
3882 xfs_fileoff_t endoffset_fsb;
3884 xfs_fsblock_t firstfsb;
3885 xfs_bmap_free_t free_list;
3886 xfs_bmbt_irec_t imap;
3887 xfs_off_t ioffset;
3891 uint resblks;
3892 uint rounding;
3894 xfs_fileoff_t startoffset_fsb;
3923 }
3930 }
3940 }
3942 /*
3943 * Need to zero the stuff we're not freeing, on disk.
3944 * If its a realtime file & can't use unwritten extents then we
3945 * actually need to zero the extent edges. Otherwise xfs_bunmapi
3946 * will take care of it for us.
3947 */
3956 xfs_daddr_t block;
3963 }
3972 mod++;
3975 }
3976 }
3978 /*
3979 * One contiguous piece to clear
3980 */
3983 /*
3984 * Some full blocks, possibly two pieces to clear
3985 */
3994 }
3996 /*
3997 * free file space until done or until there is an error
3998 */
4002 /*
4003 * allocate and setup the transaction. Allow this
4004 * transaction to dip into the reserve blocks to ensure
4005 * the freeing of the space succeeds at ENOSPC.
4006 */
4010 resblks,
4013 XFS_TRANS_PERM_LOG_RES,
4014 XFS_WRITE_LOG_COUNT);
4016 /*
4017 * check for running out of space
4018 */
4020 /*
4021 * Free the transaction structure.
4022 */
4026 }
4030 XFS_QMOPT_RES_REGBLKS);
4037 /*
4038 * issue the bunmapi() call to free the blocks
4039 */
4046 }
4048 /*
4049 * complete the transaction
4050 */
4054 }
4058 }
4060 out_unlock_iolock:
4065 error0:
4067 error1:
4070 XFS_ILOCK_EXCL);
4072 }
4074 /*
4075 * xfs_change_file_space()
4076 * This routine allocates or frees disk space for the given file.
4077 * The user specified parameters are checked for alignment and size
4078 * limitations.
4079 *
4080 * RETURNS:
4081 * 0 on success
4082 * errno on error
4083 *
4084 */
4085 int
4090 xfs_off_t offset,
4093 {
4097 xfs_fsize_t fsize;
4099 xfs_off_t startoffset;
4100 xfs_off_t llen;
4102 bhv_vattr_t va;
4120 }
4135 /*
4136 * XFS_IOC_RESVSP and XFS_IOC_UNRESVSP will reserve or unreserve
4137 * file space.
4138 * These calls do NOT zero the data space allocated to the file,
4139 * nor do they change the file size.
4140 *
4141 * XFS_IOC_ALLOCSP and XFS_IOC_FREESP will allocate and free file
4142 * space.
4143 * These calls cause the new file data to be zeroed and the file
4144 * size to be changed.
4145 */
4161 attr_flags)))
4174 }
4190 }
4192 /*
4193 * update the inode timestamp, mode, and prealloc flag bits
4194 */
4199 /* ASSERT(0); */
4202 }
4212 /*
4213 * Note that we don't have to worry about mandatory
4214 * file locking being disabled here because we only
4215 * clear the S_ISGID bit if the Group execute bit is
4216 * on, but if it was on then mandatory locking wouldn't
4217 * have been enabled.
4218 */
4223 }
4237 }